Twitch: your streams train Amazon's AI until you opt out

Twitch uses its streamers' content to train Amazon's generative AI models. The setting for it turned up in account settings on August 12, and it is switched on by default for every user. Anyone who does not want their material feeding Amazon's models has to act. There was no email about it, and no notice at login.
Generative AI models are systems that produce new text, images, audio or video from a prompt. Twitch's FAQ gives a concrete example of what your material does there. The audio from a stream may help refine speech-to-text models. That would improve captions on Twitch, but also captions across Amazon.
More than the stream is affected
According to Twitch, the setting covers your stream and the stream's chat, your VODs, your clips, highlights, and any text or images on your channel.
One sentence in the help pages matters for viewers and is easy to miss. If you type in someone else's chat, that channel's setting applies, not yours. Twitch words it like this: "If you chat on someone else's stream, their opt-out preferences govern if that chat can be used for training." So if you opt out and then post in a channel that has not, your messages still go in.
How to switch the training off
In the browser, click your profile picture at the top right, then Settings, then Security and Privacy. Scroll down to "Opt-out Preferences" and the entry "Training for Generative AI". Switch it off. The page is also reachable directly at twitch.tv/settings/security. In the mobile app the same setting sits under Profile, Settings, Security and Privacy. You only need to flip it in one place, it then applies to the account.
The wording matters. Twitch says that after an opt-out your content will not be used in future training. Whatever is already inside a model does not come back out. When a user asked in Twitch's own stream whether his videos had already been used, chief product officer Mike Minton answered that he did not know.
What the toggle does not cover
The opt-out applies to generative models only. Twitch states plainly that it does not opt anyone out of all AI or machine learning uses on the platform. AutoMod, the filter for chat messages, keeps running. Captions are still generated automatically. Recommendations and real-time sponsorship campaign assistance are also untouched.
Twitch supplies the reasoning. Some of these systems run the service itself or protect the community, so there is no individual switch for them. Minton described in the same broadcast how deep this goes. Nearly every bit of audio on the platform is transcribed into text, which is what makes stream summaries possible in the first place.
Anyone worried that opting out costs them viewers can set that aside, according to Minton. The setting has no effect on whether a channel gets discovered.
Twitch is open about the default
Minton answered the question of why the setting is pre-enabled with unusual directness: "If this was opt-in, nobody would opt in. That's honestly the answer." Almost every service in the world is on by default, he said. What is different here is that Twitch offers an opt-out at all.
The community sees it differently. On Twitch's own suggestion platform, a comment demanding opt-in collected around 14,000 upvotes within hours.
Minton's comparison does hold up in part. Google says it trains on YouTube content, including Gemini and Veo 3, and offers channel owners no way to object. Meta uses public Facebook and Instagram posts for its models as well, live streams included, also without a switch. With some other providers you can at least adjust things individually, for example when you strip the AI components out of Windows 11 or switch off the ads in ChatGPT.





